FR3082982A1 - METHOD FOR DETERMINING THE INFILTRATION OF BIOLOGICAL CELLS INTO A BIOLOGICAL OBJECT OF INTEREST - Google Patents

Abstract

The invention relates to a method for determining an infiltration profile of biological cells of interest into a biological object of interest from a digital histopathological image of biological tissues, a histological marker having previously been applied to biological tissues , comprising a generation of a biological cell detection image, pixels associated with the histological marker on the histopathological image being of a predetermined color on said image, a determination of a distance map comprising iso-curves of distance at the border of the biological object, and, from the distance map, a calculation of a curve representative of the surface density of biological cells of interest as a function of the distance from the border, by counting, for each value of distance at the border, of pixels being both of the predetermined color on the detection image, and located s between the iso-curve associated with said distance value and the consecutive iso-curve.

Description

Method for determining the infiltration of biological cells into a biological object of interest

TECHNICAL FIELD OF THE INVENTION AND STATE OF THE ART

The invention belongs to the technical field of computer processing of biological images, preferably for applications in anatomical pathology.

The diagnosis of the malignancy of a tumor, for example cancerous, and / or of the effectiveness of a treatment can be carried out by analysis of microscopic images of a sample of tumor. Slices are made in the sample taken, and these slices are treated with markers to highlight certain types of biological cells. The microscopic images of the sections, including the labeled biological cells, can be used by an anatomo-pathologist to establish the malignancy of the tumor, make a vital prognosis of the patient, evaluate the effectiveness of a treatment, etc.

In particular, the infiltration of immune cells within the tumor, for example lymphocyte infiltration, after the administration of the treatment constitutes a parameter of interest. It has been established that a high density of immune cells infiltrated within tumors can be correlated with a significant anti-tumor immune response and therefore with a good vital prognosis for patients.

A well-known solution consists in taking by biopsy coring a sample of biological tissue to be studied, then in making fine sections of said sample in order to obtain histological preparations. A health practitioner can then analyze the histological preparations under a microscope and carry out a histopathological diagnosis. However, this solution requires that the practitioner physically has the sample, which must therefore be transported and handled. In addition, the observation made by the practitioner is very little repeatable and reproducible.

Methods have been proposed for the analysis of the infiltration density of immune cells from dematerialized samples. For example, international patent application WO 2012/032173 A1 describes a method for treating a virtual tumor slide, comprising a quantification of the density of cells in a plurality of small surface areas. Said quantification zones are placed so as to overlap the tumor border. The virtual tumor slide optionally includes labeling using markers specific or not for immune cells. This document describes a preferred embodiment in which each of the quantification zones is a rectangle of small width, the perpendicular perpendicular bisector of which is superposed with the normal to the tumor front (see for example Figure 7). The quantification of cell density within an area is carried out by counting the cells in slices of predefined width of the quantification rectangle, at different distances from the tumor front. After having optionally calculated an average cell infiltration density over several quantification rectangles, a curve representative of the cell density as a function of the distance from the tumor front is obtained. This curve can be compared to typical cell infiltration profiles.

However, the method of this document does not produce an overall measurement which characterizes the infiltration of immune cells throughout the tumor. A complete analysis of the lesion is therefore not possible. The measurement only provides a general trend in the infiltration of immune cells.

In addition, known software implementing this method cannot be adapted to various types and sizes of tumors, the size of the quantification rectangles being predetermined, nor to various types of markers (in particular immunohistochemical markers). This solution is particularly difficult to use during the preclinical study phase of the efficacy of an anti-cancer treatment on non-human populations, prior to clinical trials.

GENERAL PRESENTATION OF THE INVENTION

There is therefore a need for a method making it possible to obtain an infiltration profile of a chosen type of biological cells (for example a lymphocyte infiltration profile) in a biological object of interest, whatever the type and size of the biological object.

There is in particular a need for a method providing, in a reliable and rapid manner, a global measurement of the infiltration of biological cells on the scale of the entire biological object. One challenge is to obtain a repeatable and reproducible measurement, producing objective results to assist the practitioner's decision-making, without replacing the practitioner.

An additional need exists for a method which could be used both in preclinical study, on samples taken from animals, and in clinical phase for human patients.

Furthermore, a method is preferably sought which produces satisfactory results for different types of histological markers, and for different colors of markers.

The present invention meets this need by providing, according to a first aspect, a method for determining an infiltration profile of biological cells of interest in a biological object of interest, from a digital histopathological image of tissues biological, a border of the biological object having previously been determined in the histopathological image, a histological marker having previously been applied to the biological tissues, the method, executed by a processing unit, comprising the following steps:

generation of a biological cell detection image comprising pixels of a predetermined color, said pixels corresponding to the areas of the histopathological image marked using the histological marker, determination of a distance map comprising iso-curves , each iso-curve comprising all the pixels of the region of interest located at a Euclidean distance from the border equal to a distance value;

from the distance map, calculation of a curve representative of the surface density of biological cells of interest, the calculation comprising a count, for each value of distance at the border, of pixels being both of the predetermined color on the detection image and located between the iso-curve associated with said distance value and the consecutive iso-curve.

An advantage of the method of the invention is to provide an overall measurement of the infiltration of biological cells into the biological object. The separation of the pixels of the histopathological image corresponding to the marker, associated with the determination of a distance map on the scale of the biological object, makes it possible to obtain a surface density curve of biological cells of interest at l scale of the entire biological object. The measurement is reliable and reproducible.

An additional advantage is that the results obtained by this process do not depend on the size or type of the biological object studied.

Another advantage is that this method can be implemented automatically and quickly, by a processing unit. An additional advantage is that the surface density in the context of the implementation of the method, is corrected for artifacts of the holes, folds or necrosis type, which makes the curve more robust.

Optional and non-limiting characteristics of a process of the invention are the following, considered alone or in any of their technically possible combinations:

during the calculation step, the surface density of biological cells of interest is obtained by the following formula:

where a corresponds to the number of pixels which are both of the predetermined color on the detection image and situated between the iso-curve associated with the distance value d and the consecutive iso-curve, β corresponds to the total number of pixels between the iso-curve associated with the distance value d and the consecutive iso-curve, θ is a predetermined average number of pixels per biological cell, and δ is a predetermined number of pixels per unit of area;

the step of generating the biological cell detection image comprises the following substeps:

from the histopathological image, separation of the histological marker from other markers, by image processing giving an intermediate image;

binarization of Otsu from the intermediate image, detection of biological cells of interest from the intermediate image and from an image resulting from the binarization of Otsu, by separation of a first class formed of pixels corresponding to the marker histological with respect to a second class formed by the other pixels;

- The histopathological image comprises, in addition to the histological marker, a non-specific marker, for example with hematoxylin or eosin;

- the step of separating the histological marker comprising a deconvolution of the colors of the histopathological image on the hue / saturation / luminance space;

- the histological marker is diamino-benzidine or hematoxylineaminoethylcarbazole;

the step of detecting biological cells of interest comprises a k-means classification using a centroid of the class of biological cells of interest and a centroid of the fibrosis class, the centroid of the class biological cells of interest which can then preferably be taken to be equal to the minimum gray level of the intermediate image, and the equation of the centroid of the fibrosis class can preferably be the following:

- The method comprises a preliminary step of determination, manual or automated, of the border within the histopathological image;

- the histopathological image is a virtual slide formed by tiling a plurality of microscopic images of biological tissues;

- in the distance map, the pixels of the image located outside the border are associated with a negative distance, and the pixels of the image located inside the border are associated with a positive distance, the curve of areal density of biological cells of interest including the negative distances on a left part and the positive distances on a right part;

- the method includes an additional step of graphical representation of an infiltration map of the cells of interest on which the iso-curves are superimposed on the histopathological image of the region of interest, the color of an area between two consecutive iso-curves being variable as a function of the surface density value of biological cells of interest;

the method comprises additional steps of determining an area under the surface density curve) of biological cells of interest as a function of the distance at the border, for a given interval of distance, and of comparing the area obtained at a predetermined threshold value;

- the biological object of interest is a tumor or a set of tumors, and the border is a tumor border;

- the histopathological image, on which the border has previously been determined, is downloaded by the processing unit to a remote database, preferably via a network connection and even more preferably by via an Internet connection;

the biological cells of interest are, for example and without limitation, T lymphocytes, or B lymphocytes, or NK cells.

The present invention relates, according to a second aspect, to a processing unit comprising:

a memory configured to record a histopathological image, a first sub-unit configured to determine a distance map comprising iso-curves, each iso-curve being associated with a distance value and comprising all the pixels of the image located at a Euclidean distance from a border equal to said distance value, a second subunit configured to calculate a curve representative of the surface density of biological cells of interest in the image as a function of the distance from the border , the processing unit being configured to implement the method as defined above.

Finally, the invention relates to a computer program product comprising code instructions which, when implemented by a processing unit, allow the execution of a method as defined above.

GENERAL PRESENTATION OF THE FIGURES

Other characteristics, objects and advantages of the invention will emerge from the description which follows, which is purely illustrative and not limiting, accompanied by the appended drawings, among which:

Figure 1 schematically shows a system for implementing a method for determining an infiltration profile of biological cells of interest;

FIG. 2 represents the steps of a method according to an embodiment of the invention, in which the biological object studied is a tumor;

Figure 3 is a first image from a virtual slide of a section of biological tissue comprising a tumor;

Figure 4a is a second image from a virtual slide of a section of biological tissue comprising a tumor, in which the lymphocytes have been labeled;

Figure 4b is a grayscale image from the separation of the DAB marker in the image of Figure 4a;

Figure 4c is a gray level image from the separation of the hematoxylin marker in the image of Figure 4a;

Figure 5 is a gray level image from the separation of the immunohistochemical marker in the image of Figure 3;

Figure 6 is an image from Otsu's binarization of the image with separate markers in Figure 5;

Figure 7 is a lymphocyte detection image obtained from the image of Figure 3, where the pixels corresponding to lymphocytes are in black;

Figure 8 is an image comprising, in superposition, the image of Figure 5 on the one hand, and iso-curves of a distance map to the tumor border identified in Figure 7 on the other hand; there is shown at the bottom of Figure 7b a close-up view of an area of this image;

FIG. 9 is a diagram representing the distribution of a lymphocyte between several values of distance from the tumor border;

Figure 10 is an example of a lymphocyte infiltration profile obtained according to the method of Figure 2.

DETAILED DESCRIPTION

A method for determining an infiltration profile and an associated system are described below, for the particular case where the biological objects are tumors, for example cancerous tumors. The border of the biological object is therefore a tumor border in the examples below. It will however be understood that the invention can be used with the same advantages for the study of the infiltration of biological cells into biological objects other than tumors. The following method can for example be applied to determine measures of infiltration of immune cells (such as lymphocytes) into a gland of the human or animal body, or into an organ transplanted to a patient, to determine whether there is rejection of the transplanted organ.

Subsequently, by “histopathological image” is meant an image allowing a microscopic study of biological tissues, usable in histopathology in particular for monitoring a disease. The biological tissues visible on such an image are most often biopsies or tissues removed during a surgical operation. The methods of preparing a sample in order to obtain a histopathological image of satisfactory quality are well known to those skilled in the art, and will not be described below. One image or a plurality of histopathological images can for example be used to establish a diagnosis of a human or animal patient, or to evaluate the effectiveness of a treatment administered to said patient. The histopathological images described below comprise at least part of a tumor border, that is to say a line of demarcation identifiable on the histopathological image between tumor tissue and non-tumor tissue. However, in some cases, isolated tumor cells may be present beyond the tumor front, especially if there is a phenomenon of "tumor budding".

Biological cell infiltration profit determination system

FIG. 1 shows a system allowing, once a sample of biological tissue comprising a biological object of interest such as a tumor has been taken, the implementation of the method for determining the lymphocyte infiltration profile which will be described below.

The system comprises a processing unit 10 comprising calculation means, preferably a processor. The processing unit 10 may further include a user interface to allow entry of instructions by a user. The processing unit 10 includes a memory 11 configured for saving histopathological images, possibly annotated to indicate a tumor border, and for saving lymphocyte infiltration profiles determined using the method which will be described below. . The memory 11 can, optionally, also be configured to save the intermediate images generated during the execution of the method. The processing unit is advantageously connected with or without wire to a screen 12 comprising a graphical interface. The screen 12 is notably configured to display histopathological images.

In a possible variant, the system comprises a histopathological image acquisition unit 13, configured to generate histopathological images from real biological tissues, the histopathological images preferably being stored in the form of virtual slides. By “virtual slide” is meant an image of biological tissues recomposed from several microscopic images, for example microscopic images at several different resolutions, for example by tiling. The virtual slides can in particular be WSI (“Whole Slide Image”) type images comprising very high resolution views of tissue biopsies.

Preferably, the histopathological image acquisition unit 13 is a full-blade scanner. Alternatively, the unit 13 can be an optical microscope equipped with a camera.

Alternatively or in combination, the system comprises a remote database 14 from a remote server, the database 14 being configured to store histopathological images, preferably virtual slides, in memory. The database 14 is able to communicate with the processing unit 10 via a communication network R, preferably an Internet network, the processing unit 10 being configured in this variant to download the histopathological images via the network R.

Preferably, virtual slides of biological tissues of individuals are shared over the secure network R, preferably with secure access to the data. The practitioner can thus, using known image processing software, obtain a view at a more or less high resolution of different zones of a region of interest comprising a biological object to be studied. Such virtual slides and methods for their acquisition and sharing are well known to those skilled in the art. The virtual slide is made up of a plurality of slices of a tumor sample, placed on slides and treated with histological markers. According to a preferred embodiment, the virtual blade consists of a single slice.

The use of virtual slides is very advantageous, since these slides can very easily be shared and annotated on known image processing software, remotely and without the need to manipulate a physical sample of the biological tissues to be analyzed.

In addition, such a virtual slide has the advantage, compared to a biological sample, of not degrading over time.

Finally, it is easier for a user to use a virtual slide of a biological object, rather than a plurality of different views of the same biological object separated from each other.

Method for determining a lymphocyte infiltration profit in a tumor

FIG. 2 represents the steps of a method of the invention according to a particular embodiment, in which the biological object to be studied is a cancerous tumor, the border of said object is a tumor border (or tumor front) and the cells biological of interest whose infiltration is to be quantified are lymphocytes such as T lymphocytes or B lymphocytes.

Note that the method of Figure 2 can be applied with the same advantages to quantify the infiltration of other types of biological cells, including NK cells, macrophages, dendritic cells.

For purely illustrative purposes, the method of FIG. 2 can be used to evaluate, in a preclinical study of a treatment for colorectal cancer (or CRC), the response of a tumor to a radiofrequency treatment followed by an injection of 'a GMCF gel (for “Gramulocyte Macrophage Colony Stimulating Factor”) having the role of activating the dendritic cells responsible for triggering the immune response. The aim of such treatment is to destroy tumor cells which may recur after radiofrequency and to limit the appearance of other tumor cells distant from the treatment site. The tumor being studied may be a primary tumor or a secondary metastasis.

The method of FIG. 2 makes it possible to obtain a profile of infiltration of lymphocytes into the tumor, that is to say a curve of quantification of the density of lymphocytes as a function of the distance from the tumor front. This method can in particular be implemented by the processing unit 10 of the system shown diagrammatically in FIG. 1.

The method of FIG. 2 takes as input a histopathological image I or a plurality of histopathological images of a tumor, preferably a virtual slide of tumor.

The tumor has been previously marked using at least one type of histological marker. The expression “histological marker” refers to a chemical marker used in histology, allowing labeling of biological tissues. A distinction is made between non-specific markers, marking all the cells of biological tissues in the same way, such as hematoxylin which stains the DNA contained in the nuclei in blue-violet or eosin which stains the cytoplasm proteins in pink, and specific markers which mark only certain cells of interest. The specific markers can in particular be immunohistochemical markers (or IHC markers), operating on the principle of a binding reaction between an antibody (present in the marker) and an antigen. Such a marker is for example coupled to an enzyme which produces a specific coloration.

The IHC markings can be of direct or indirect type. For direct labeling, a visual marker (or an enzyme) is coupled to an antibody capable of binding directly with the antigen which it is desired to label. The indirect method is carried out in two stages. First, an untagged primary antibody (without visual marker) is linked to the targeted antigen; second, after rinsing off the excess, a secondary antibody with visual marker (or an enzyme) is added and binds to the primary antibody.

Diamino-benzidine (or DAB, brown coloring) and hematoxylineaminoethylcarbazole (or H-AEC, blue coloring for the nucleus and rougemarron for the outline) are known immunohistochemical markers. The observation of the images resulting from an IHC labeling generally gives halos tinted with the tint of the marker, on the periphery of the biological cells of interest targeted by the labeling, and nuclei tinted with the tint of the counter-marker (blue if the counter-marker is hematoxylin).

In the example which follows, the histopathological image is marked with diamino-benzidine (DAB below) and hematoxylin (as an alternative or in addition to hematoxylin, the image may include a labeling with eosin).

Preliminary, a histopathological image of the tumor, including an indication of the tumor border (said border is preferably visible using an annotation of the image, produced by a practitioner) is necessary. As such, the method of FIG. 2 includes, optionally, a step 100 of determining the tumor border F. The border F, which delimits the biological object of interest constituted by the tumor (or possibly a set of tumors), here is a closed curve. Alternatively, the border F can be an open curve.

Step 100 is implemented either automatically, by automatic reading of the image I marked using a histological marker, or manually by an operator, preferably a health practitioner trained in histopathology.

Alternatively, the histopathological image I received by the processing unit implementing the method may already include an annotation of the tumor border F. For example, if the image I is acquired by the processing unit via the Internet from d '' a sharing platform, image I may have been previously annotated by a practitioner, using image processing software, to reveal the tumor border F.

Step 100 makes it possible to identify the region of interest comprising the biological object to be studied (here a tumor) for the following steps. The annotation 100 of the border F advantageously replaces an automatic segmentation step of the region of interest comprising the tumor.

FIG. 3 shows a histopathological image I derived from a virtual slide of a tumor, and showing an annotation of the tumor border F. In this example, the biological object to be studied (ie the tumor ), located inside the closed curve F, corresponds to a surface having a lower brightness compared to the rest of the image. The tumor border F is indicated using a broken curve in thick line. The histopathological image I, stored temporarily or permanently by the processing unit 10, is preferably a large image, for example more than 3 gigabytes and a resolution of 150,000 x 100,000 pixels. Image I is for example an image of the WSI (Whole Slide Image) type. In image I, the tumor delimited by the border F has a diameter of approximately 2 millimeters.

The method for determining the lymphocyte infiltration profile continues with a step 200 of detecting pixels of image I corresponding to lymphocytes.

It is recalled that the biological tissues comprising the tumor were marked using a histological marker, preferably an immunohistochemical marker specific for lymphocytes, before the acquisition of image I. The detection of lymphocytes exploits the distinct marking lymphocytes in image I, compared to the rest of the elements visible in the image.

The histopathological image of the present example includes DAB as a specific marker for lymphocytes and hematoxylin as a countermarker. Initially, a sub-step 210 of separation of these two markers on the image is implemented.

A known method in histology for the separation of markers is based on a color deconvolution, as a function of the absorption of light by the different markers (Beer-Lambert law, see Ruifrok, Johnston et al. (2001), Quantification of histochemical staining by color deconvolution, Analytical and quantitative cytology and histology, 23 (4): 291-299). However, the immunohistochemistry markers disperse the incident light, instead of absorbing it, and BeerLambert's law is therefore not applicable.

It has been found that the deconvolution of the colors of the pixels of image I in the hue-saturation-luminance space or “Hue-SaturationLightness” (HSL space below) makes it possible to effectively discriminate the different markers, in particular immunohistochemicals, on a picture. Step 210 takes as input a hue, a saturation and a reference luminance of the immunohistochemical marker sought, making it possible to place a reference point of said marker in the HSL space in three dimensions. During this step, an intermediate lint image separating the markers is generated, each pixel of the lint image having a gray level which corresponds to the distance between the corresponding pixel of the image I and the reference point in the HSL area. The closer a pixel of the image I is to the reference point in the HSL space, the darker the corresponding pixel of the lint image.

For example, if the reference point has the coordinates (ho, so, Io) in the HSL space, an equation of the distance I (color space function) of a pixel p with coordinates (h, s , I) in the HSL space with respect to said reference point is as follows:

</// (/)))

with the following values of dn (p), ds (p) and di_ (p):

// and a value of the parameter ε which can be fixed at 1/255.

By way of illustration, FIG. 4a shows a histopathological image obtained from a virtual slide of another tumor sample than that represented in FIG. 3. As with the image of FIG. 3, the tumor of FIG. 4a was labeled with DAB and hematoxylin. The tumor border is clearly visible and is not shown.

Figure 4b is an image obtained by separating the markers according to the distance calculation above, considering a reference point associated with the DAB. The dark pixels of the image of FIG. 4b therefore correspond to lymphocytes on the image before separation of the markers. This image is of interest for the quantification of lymphocyte infiltration.

The image of Figure 4c is obtained by separating the markers according to the same method, by considering a reference point associated with hematoxylin. The dark pixels in the image in Figure 4c correspond to DNA.

FIG. 5 shows the intermediate lint image of separation of the markers resulting from the histopathological image I (without border F) of FIG. 3, with separation of the markers in the HSL space. The lint image used for the rest of the process is obtained with a reference point having a predetermined hue, saturation and luminance corresponding to the DAB marker. A dotted zone Z has been indicated in FIG. 5, corresponding to a bluish zone of the histopathological image I, resulting from the labeling with hematoxylin; it can be seen that the image separating the markers Imt does not make it possible to distinguish this bluish zone with respect to other white pixels in the image.

Alternatively, the intermediate lint image can be generated using the distance to a reference point corresponding to the marker in red-green-blue or "Red Green Blue" (RGB) space.

Step 200 of the method then comprises an automated detection of the pixels corresponding to lymphocytes on image I. In the present example, the detection of lymphocytes results in a binary image, where the pixels corresponding to lymphocytes are of a predetermined color (black), and all other pixels are the background color (white).

From the lint image of separation of the specific marker for lymphocytes, the processing unit firstly implements a binarization 220, for example a binarization according to the Otsu method. Binarization produces, from the lint image in gray levels, a threshold image I '. The threshold separating the class of foreground pixels (including the pixels to be detected) and the class of background pixels is optimized during Otsu binarization so as to obtain a minimum intra-class variance. FIG. 6 shows the thresholded image I ′ obtained after binarization of Otsu from the intermediate image of FIG. 5.

The black pixels in the image of Figure 6, detected as foreground pixels, include both pixels associated with lymphocytes and pixels associated with fibrosis.

Step 200 according to the present example comprises an additional step 230 of detecting lymphocytes, in particular in order to distinguish lymphocytes from fibrosis. One objective is to separate, on image I, a first class formed of pixels corresponding to the DAB marker, with respect to a second class formed of the other pixels of the image.

Step 230 therefore includes a “k-means” classification of the pixels between two classes, in other words two “clusters”, from the intermediate image lint and from the thresholded image I '. For the application of the k-means algorithm, the centroid of the lymphocyte class is advantageously initialized to the minimum of the gray levels detected in the intermediate image lint, and the centroid of the fibrosis class is initialized to a value a. The value a is advantageously obtained using the following equation:

where the summing is carried out on the pixels p of the intermediate image lint of separation of the markers, and of the image I 'obtained by binarization. The step 230 of detecting lymphocytes by k-means classification is advantageous, because it gives results of good precision for the detection of the objects of interest, here the lymphocytes. Because the histopathological images studied are two-dimensional images derived from real biological samples (three-dimensional), it is generally difficult to distinguish lymphocytes from other types of cells in the sample; however, the results obtained by kmeans classification are satisfactory.

Figure 7 represents the image I * of detection of lymphocytes resulting from the k-means classification applied to the binary image I '. Image I *, which is a binary image, contains in black the pixels corresponding to the lymphocytes visible in image I, and in white the other pixels. Image I * can therefore be used for the calculation of the density of lymphocytes.

It will be noted that, alternatively, the step 200 of detection of biological cells of interest can be carried out before the annotation 100 of the tumor border F, or in parallel with the annotation. It is not necessary to have the position of the F border to detect the cells highlighted by the histological marker.

The infiltration profile determination method then comprises, in an important manner, the determination 300 of a distance map DM.

An advantage of using a distance map with respect to the tumor border is to allow quantification of the infiltration of lymphocytes, as a function of their distance from the edge of the tumor.

Advantageously, a detection of an area inside the tumor and of an area outside the tumor is implemented, using Jordan's theorem. A region of interest (or ROI) may have been previously identified in image I.

Then, from the histopathological image I, the Euclidean distance of each of the pixels of the region of interest relative to the tumor border F, previously annotated or detected, is calculated. The distance of a pixel, with respect to the border F, is defined as the minimum of the set formed by the Euclidean distances between said pixel and each of the pixels of the border F.

Once the said distances have been calculated, successive thresholds of the pixels of the image I are made as a function of their distance from the border F, so as to obtain a set of iso-curves. An iso-curve DMd of the distance map DM is defined as the curve connecting all the pixels located at a distance from the border F equal to the value d. A representation of the distance map DM can be obtained by superimposing the iso-curves DMd on the histopathological image I.

FIG. 8 represents the distance map DM obtained at the end of step 300, for the image I of FIG. 3. By convention, the pixels outside the border F are defined as having a negative distance at the border F, and the pixels inside the border are defined as having a positive distance from the border. The DMd iso-curves corresponding to positive distances have been shown. A rectangle of image I has been represented on a larger scale at the bottom of FIG. 8. The iso-curve DMo corresponds to the pixels neighboring the border F. We have also identified the iso-curve DMs, connecting the pixels located at a distance d = 5 pm from the border F.

In Figure 8, the distance map is obtained at the scale of the whole tumor. Determining the distance map does not require placing quantification areas around the tumor.

It will be noted that step 300 of determining the distance map DM can, alternatively, be implemented prior to the detection of lymphocytes in step 200, or in parallel with the detection of lymphocytes. The calculation of the distance map DM takes as input the histopathological image I.

The distance map DM on the one hand, and the image I * of detection of lymphocytes on the other hand, can be used jointly to group together the pixels detected as corresponding to lymphocytes according to their distance from the tumor front. As such, the method of FIG. 2 comprises a step 400 of determining the profile of infiltration of the lymphocytes as a function of their distance from the border F. The step 400 takes as input the distance map DM, the histopathological image I and the binary image I * for detecting lymphocytes.

For each value of distance d from the border F, a density f (d) of lymphocytes per unit area corresponding to said distance is determined in step 400. A number δ of pixels per unit area is predetermined as a function of the resolution of image I, as well as an average number pixels of pixels per lymphocyte. We therefore reason with an average size of lymphocyte, which makes it possible to obtain a general tendency of lymphocyte infiltration. For an image resolution of 0.329 micrometers per pixel, there is for example a δ value of 46193.2167 pixels per square micrometer, and an θ value of 315 pixels per lymphocyte on average. By having these predetermined values, the surface density of lymphocytes is obtained as follows during a calculation sub-step 420:

_ei . ^ a * δ ⁼ (a + b) * Θ

The value a corresponds to the number of pixels detected as marked at the DAB, among the pixels located at a distance d from the border F, while the value b corresponds to the difference in the total number of pixels located at the distance d with the number a of pixels marked at DAB. The number (a + b) is therefore the total number of pixels located at the distance d from the edge F. The value a is obtained in a sub-step 410 during which, for the value d of distance, are counted the pixels being both black on image I * (therefore detected as marked with DAB on image I) and located between the iso-curve DMd and the consecutive iso-curve DMd + i.

Note that counting 410 is carried out only as a function of the black or white color of the pixels of the image I *. The number of lymphocytes crossing the DMd iso-curve is not taken into account. FIG. 9 shows a superimposed view, on a larger scale than that of FIG. 3, of image I and the DMd iso-curves, on which two Li and L2 lymphocytes have been identified; the pixels corresponding to the Li and L2 lymphocytes are distributed, during step 410, between several distance values, since they overlap iso-curves. For example, the Li lymphocyte extends between the DMd and DMd + 2 iso-curves; about 40% of the pixels corresponding to Li are counted as being at the distance d from the tumor border, and about 60% of the pixels corresponding to Li are counted as being at the distance d + 1.

The measurement of density of lymphocytes per unit area is thus obtained on the scale of an entire tumor. This measurement is repeatable and reproducible, in particular because it does not depend on the placement of quantification zones; this measurement mainly depends on the location of the F border, and on the values of the predetermined parameters for binarization and for counting the pixels corresponding to ds lymphocytes.

Optionally, the surface density values of lymphocytes obtained at the end of step 400 are represented on a curve as illustrated in FIG. 10, showing the values of distance to the tumor border on the abscissa, and on the ordinate the surface density of lymphocytes for each of these distances. By convention, negative distances (outside the tumor border) are shown on the left and positive distances (inside the tumor border) are shown on the right.

The curve thus obtained provides a lymphocyte infiltration profile within the tumor visible on the histopathological image I.

Optionally, observation of the infiltration profile by an expert, for example an anatomo-pathologist, allows the classification of the profile in a category, among the following four types of lymphocyte infiltration, associated with different clinical results (see as such Allard et al. (2012), Linear quantification of lymphoid infiltration of the tumor margin: a reproducible method, developed with colorectal cancer tissues, for assessing highly variable prognostic factor, Diagnostic Pathology, 7 (1): 156, as well as Emile et al. (2017), Histological classification and molecular alterations of histiocytosis, La Presse Médicale, 46 (1): 4654):

- Type 1: No lymphocytic infiltration;

- Type 2: Low lymphecytic infiltration of the tumorous frent;

- Type 3: Fprte lymphocyte infiltration of the tumor front only;

- Type 4: Fprte lymphecytic infiltration of the tumor front and the interior of the tumor.

Indeed, a lymphecytic infiltration profile obtained using the method described above makes it possible to distinguish between the case of an infiltration of the tumor front only (high surface density of lymphccytes only fear of low values of distance to the border), and the case of lymphecytic infiltration to the heart of the tumor (surface density of lymphocytes also high for distances far from the border).

Claims (16)

1. Method for determining an infiltration profile of biological cells of interest in a biological object of interest, from a histopathological image (I) digital of biological tissues, a border (F) of the object biological having previously been determined within the histopathological image, a histological marker having previously been applied to the biological tissues, the method, executed by a processing unit, comprising the generation (200) of a detection image (I *) biological cells (Γ) comprising pixels of a predetermined color, the pixels of the predetermined color corresponding to areas of the histopathological image (I) marked using the histological marker, the method being characterized in that it also includes the following steps: - determination (300) of a distance map (DM) comprising iso-curves (DMd), each iso-curve comprising all the pixels (x, y) of the region of interest located at a Euclidean distance ( D (x, y)) at the border (F) equal to a distance value (d); - from the distance map (DM), calculation (400) of a curve representative of the surface density (f (d)) of biological cells of interest, the calculation comprising a count (410), for each value of distance (d), of pixels being both of the predetermined color on the detection image (I *) and located between the iso-curve (DMd) associated with said distance value (d) and the iso - consecutive curve. 2. Method according to claim 1, in which during the calculation step (400), the surface density (f (d)) of biological cells of interest is obtained by the following formula: where a corresponds to the number of pixels which are both of the predetermined color on the detection image (I *) and located between the iso-curve (DMd) associated with the distance value d and the iso-curve consecutive, β corresponds to the total number of pixels between the iso-curve (DMd) associated with the distance value d and the consecutive iso-curve, θ is a predetermined average number of pixels per biological cell, and δ is a number predetermined number of pixels per unit area. 3. Method according to claim 1 or 2, in which the step (200) of generating the image (I *) of detection of biological cells comprises the following substeps: -from the histopathological image (I), separation (210) of the histological marker from other markers, by image processing giving an intermediate image (lint); - binarization of Otsu (220) of the intermediate image (lint), -detection (230) of biological cells of interest from the intermediate image (Lnt) and from an image resulting from the binarization of Otsu, by separation of a first class formed of pixels corresponding to the histological marker relative to a second class formed by the other pixels. 4. Method according to claim 3, in which the histopathological image (I) comprises, in addition to the histological marker, a non-specific marker, for example with hematoxylin or eosin, the separation step (210) of the histological marker comprising a deconvolution of the colors of the histopathological image (I) on the hue / saturation / luminance space. 5. Method according to one of claims 3 or 4, wherein the step of detecting (230) biological cells of interest comprises a k-means classification using a centroid of the class of biological cells d interest and a centroid of the fibrosis class. 6. Method according to claim 5, in which the centroid of the class of biological cells of interest is taken equal to the minimum gray level of the intermediate image (lint), and in which the equation of the centroid of the fibrosis dasse is the following :

7. Method according to one of claims 1 to 6, comprising a preliminary step of determination (100), manual or automated, of the border (F) within the histopathological image (I). 8. Method according to one of claims 1 to 7, wherein the histopathological image (I) is a virtual slide formed by tiling a plurality of microscopic images of biological tissues. 9. Method according to one of claims 1 to 8, where in the distance map (DM), the pixels of the image located outside the border (F) are associated with a negative distance (d), and the pixels of the image located inside the border (F) are associated with a positive distance (d), the surface density curve (f (d)) of biological cells of interest comprising the negative distances on a left part and the positive distances on a straight section. 10. Method according to one of claims 1 to 9, comprising an additional step of graphical representation (500) of an infiltration map of the cells of interest (C) on which the iso-curves (DMd) are superimposed on the histopathological image (I) of the region of interest, the color of an area between two consecutive iso-curves (DMd) being variable as a function of the surface density value (f (d)) of biological cells of interest. 11. Method according to one of claims 1 to 10, comprising additional steps of determining an area under the surface density curve (f (d)) of biological cells of interest as a function of the distance at the border ( F), for a given interval of distance, and of comparison of the area obtained with a predetermined threshold value. 12. Method according to one of claims 1 to 11, wherein the biological object of interest is a tumor or a set of tumors, and the border (F) is a tumor border. 13. Method according to one of claims 1 to 12, in which the histopathological image (I), on which the border (F) has previously been determined, is downloaded by the processing unit from a database (12) remote, preferably via a network connection. 14. Method according to one of claims 1 to 13, wherein the biological cells of interest are T lymphocytes, or B lymphocytes, or NK cells. 15. Processing unit (10) comprising: - a memory (11) configured to record a histopathological image (I), a first sub-unit configured to determine a distance map (DM) comprising iso-curves (DMd), each iso-curve being associated with a distance value (d) and comprising all of the pixels (x, y ) of the histopathological image (I) located at a Euclidean distance (D (x, y)) from a border (F) equal to said distance value (d), - a second subunit configured to calculate a curve representative of the surface density of biological cells of interest in the histopathological image (I) as a function of the distance from the border (F), the processing unit being configured to implement the method according to one of claims 1 to 14. 16. Product computer program including code instructions 5 which, when executed by a processing unit, allow the implementation of the method for determining the infiltration profile of one of claims 1 to 14.